2 * offload engine driver for the Intel Xscale series of i/o processors
3 * Copyright © 2006, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
21 * This driver supports the asynchrounous DMA copy and RAID engines available
22 * on the Intel Xscale(R) family of I/O Processors (IOP 32x, 33x, 134x)
25 #include <linux/init.h>
26 #include <linux/module.h>
27 #include <linux/delay.h>
28 #include <linux/dma-mapping.h>
29 #include <linux/spinlock.h>
30 #include <linux/interrupt.h>
31 #include <linux/platform_device.h>
32 #include <linux/memory.h>
33 #include <linux/ioport.h>
34 #include <linux/raid/pq.h>
35 #include <linux/slab.h>
37 #include <mach/adma.h>
39 #define to_iop_adma_chan(chan) container_of(chan, struct iop_adma_chan, common)
40 #define to_iop_adma_device(dev) \
41 container_of(dev, struct iop_adma_device, common)
42 #define tx_to_iop_adma_slot(tx) \
43 container_of(tx, struct iop_adma_desc_slot, async_tx)
46 * iop_adma_free_slots - flags descriptor slots for reuse
48 * Caller must hold &iop_chan->lock while calling this function
50 static void iop_adma_free_slots(struct iop_adma_desc_slot
*slot
)
52 int stride
= slot
->slots_per_op
;
55 slot
->slots_per_op
= 0;
56 slot
= list_entry(slot
->slot_node
.next
,
57 struct iop_adma_desc_slot
,
63 iop_desc_unmap(struct iop_adma_chan
*iop_chan
, struct iop_adma_desc_slot
*desc
)
65 struct dma_async_tx_descriptor
*tx
= &desc
->async_tx
;
66 struct iop_adma_desc_slot
*unmap
= desc
->group_head
;
67 struct device
*dev
= &iop_chan
->device
->pdev
->dev
;
68 u32 len
= unmap
->unmap_len
;
69 enum dma_ctrl_flags flags
= tx
->flags
;
74 src_cnt
= unmap
->unmap_src_cnt
;
75 dest
= iop_desc_get_dest_addr(unmap
, iop_chan
);
76 if (!(flags
& DMA_COMPL_SKIP_DEST_UNMAP
)) {
77 enum dma_data_direction dir
;
79 if (src_cnt
> 1) /* is xor? */
80 dir
= DMA_BIDIRECTIONAL
;
82 dir
= DMA_FROM_DEVICE
;
84 dma_unmap_page(dev
, dest
, len
, dir
);
87 if (!(flags
& DMA_COMPL_SKIP_SRC_UNMAP
)) {
89 addr
= iop_desc_get_src_addr(unmap
, iop_chan
, src_cnt
);
92 dma_unmap_page(dev
, addr
, len
, DMA_TO_DEVICE
);
95 desc
->group_head
= NULL
;
99 iop_desc_unmap_pq(struct iop_adma_chan
*iop_chan
, struct iop_adma_desc_slot
*desc
)
101 struct dma_async_tx_descriptor
*tx
= &desc
->async_tx
;
102 struct iop_adma_desc_slot
*unmap
= desc
->group_head
;
103 struct device
*dev
= &iop_chan
->device
->pdev
->dev
;
104 u32 len
= unmap
->unmap_len
;
105 enum dma_ctrl_flags flags
= tx
->flags
;
106 u32 src_cnt
= unmap
->unmap_src_cnt
;
107 dma_addr_t pdest
= iop_desc_get_dest_addr(unmap
, iop_chan
);
108 dma_addr_t qdest
= iop_desc_get_qdest_addr(unmap
, iop_chan
);
111 if (tx
->flags
& DMA_PREP_CONTINUE
)
114 if (!(flags
& DMA_COMPL_SKIP_DEST_UNMAP
) && !desc
->pq_check_result
) {
115 dma_unmap_page(dev
, pdest
, len
, DMA_BIDIRECTIONAL
);
116 dma_unmap_page(dev
, qdest
, len
, DMA_BIDIRECTIONAL
);
119 if (!(flags
& DMA_COMPL_SKIP_SRC_UNMAP
)) {
122 for (i
= 0; i
< src_cnt
; i
++) {
123 addr
= iop_desc_get_src_addr(unmap
, iop_chan
, i
);
124 dma_unmap_page(dev
, addr
, len
, DMA_TO_DEVICE
);
126 if (desc
->pq_check_result
) {
127 dma_unmap_page(dev
, pdest
, len
, DMA_TO_DEVICE
);
128 dma_unmap_page(dev
, qdest
, len
, DMA_TO_DEVICE
);
132 desc
->group_head
= NULL
;
137 iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot
*desc
,
138 struct iop_adma_chan
*iop_chan
, dma_cookie_t cookie
)
140 struct dma_async_tx_descriptor
*tx
= &desc
->async_tx
;
142 BUG_ON(tx
->cookie
< 0);
143 if (tx
->cookie
> 0) {
147 /* call the callback (must not sleep or submit new
148 * operations to this channel)
151 tx
->callback(tx
->callback_param
);
153 /* unmap dma addresses
154 * (unmap_single vs unmap_page?)
156 if (desc
->group_head
&& desc
->unmap_len
) {
157 if (iop_desc_is_pq(desc
))
158 iop_desc_unmap_pq(iop_chan
, desc
);
160 iop_desc_unmap(iop_chan
, desc
);
164 /* run dependent operations */
165 dma_run_dependencies(tx
);
171 iop_adma_clean_slot(struct iop_adma_desc_slot
*desc
,
172 struct iop_adma_chan
*iop_chan
)
174 /* the client is allowed to attach dependent operations
177 if (!async_tx_test_ack(&desc
->async_tx
))
180 /* leave the last descriptor in the chain
181 * so we can append to it
183 if (desc
->chain_node
.next
== &iop_chan
->chain
)
186 dev_dbg(iop_chan
->device
->common
.dev
,
187 "\tfree slot: %d slots_per_op: %d\n",
188 desc
->idx
, desc
->slots_per_op
);
190 list_del(&desc
->chain_node
);
191 iop_adma_free_slots(desc
);
196 static void __iop_adma_slot_cleanup(struct iop_adma_chan
*iop_chan
)
198 struct iop_adma_desc_slot
*iter
, *_iter
, *grp_start
= NULL
;
199 dma_cookie_t cookie
= 0;
200 u32 current_desc
= iop_chan_get_current_descriptor(iop_chan
);
201 int busy
= iop_chan_is_busy(iop_chan
);
202 int seen_current
= 0, slot_cnt
= 0, slots_per_op
= 0;
204 dev_dbg(iop_chan
->device
->common
.dev
, "%s\n", __func__
);
205 /* free completed slots from the chain starting with
206 * the oldest descriptor
208 list_for_each_entry_safe(iter
, _iter
, &iop_chan
->chain
,
210 pr_debug("\tcookie: %d slot: %d busy: %d "
211 "this_desc: %#x next_desc: %#x ack: %d\n",
212 iter
->async_tx
.cookie
, iter
->idx
, busy
,
213 iter
->async_tx
.phys
, iop_desc_get_next_desc(iter
),
214 async_tx_test_ack(&iter
->async_tx
));
216 prefetch(&_iter
->async_tx
);
218 /* do not advance past the current descriptor loaded into the
219 * hardware channel, subsequent descriptors are either in
220 * process or have not been submitted
225 /* stop the search if we reach the current descriptor and the
226 * channel is busy, or if it appears that the current descriptor
227 * needs to be re-read (i.e. has been appended to)
229 if (iter
->async_tx
.phys
== current_desc
) {
230 BUG_ON(seen_current
++);
231 if (busy
|| iop_desc_get_next_desc(iter
))
235 /* detect the start of a group transaction */
236 if (!slot_cnt
&& !slots_per_op
) {
237 slot_cnt
= iter
->slot_cnt
;
238 slots_per_op
= iter
->slots_per_op
;
239 if (slot_cnt
<= slots_per_op
) {
246 pr_debug("\tgroup++\n");
249 slot_cnt
-= slots_per_op
;
252 /* all the members of a group are complete */
253 if (slots_per_op
!= 0 && slot_cnt
== 0) {
254 struct iop_adma_desc_slot
*grp_iter
, *_grp_iter
;
255 int end_of_chain
= 0;
256 pr_debug("\tgroup end\n");
258 /* collect the total results */
259 if (grp_start
->xor_check_result
) {
260 u32 zero_sum_result
= 0;
261 slot_cnt
= grp_start
->slot_cnt
;
262 grp_iter
= grp_start
;
264 list_for_each_entry_from(grp_iter
,
265 &iop_chan
->chain
, chain_node
) {
267 iop_desc_get_zero_result(grp_iter
);
268 pr_debug("\titer%d result: %d\n",
269 grp_iter
->idx
, zero_sum_result
);
270 slot_cnt
-= slots_per_op
;
274 pr_debug("\tgrp_start->xor_check_result: %p\n",
275 grp_start
->xor_check_result
);
276 *grp_start
->xor_check_result
= zero_sum_result
;
279 /* clean up the group */
280 slot_cnt
= grp_start
->slot_cnt
;
281 grp_iter
= grp_start
;
282 list_for_each_entry_safe_from(grp_iter
, _grp_iter
,
283 &iop_chan
->chain
, chain_node
) {
284 cookie
= iop_adma_run_tx_complete_actions(
285 grp_iter
, iop_chan
, cookie
);
287 slot_cnt
-= slots_per_op
;
288 end_of_chain
= iop_adma_clean_slot(grp_iter
,
291 if (slot_cnt
== 0 || end_of_chain
)
295 /* the group should be complete at this point */
304 } else if (slots_per_op
) /* wait for group completion */
307 /* write back zero sum results (single descriptor case) */
308 if (iter
->xor_check_result
&& iter
->async_tx
.cookie
)
309 *iter
->xor_check_result
=
310 iop_desc_get_zero_result(iter
);
312 cookie
= iop_adma_run_tx_complete_actions(
313 iter
, iop_chan
, cookie
);
315 if (iop_adma_clean_slot(iter
, iop_chan
))
320 iop_chan
->completed_cookie
= cookie
;
321 pr_debug("\tcompleted cookie %d\n", cookie
);
326 iop_adma_slot_cleanup(struct iop_adma_chan
*iop_chan
)
328 spin_lock_bh(&iop_chan
->lock
);
329 __iop_adma_slot_cleanup(iop_chan
);
330 spin_unlock_bh(&iop_chan
->lock
);
333 static void iop_adma_tasklet(unsigned long data
)
335 struct iop_adma_chan
*iop_chan
= (struct iop_adma_chan
*) data
;
337 /* lockdep will flag depedency submissions as potentially
338 * recursive locking, this is not the case as a dependency
339 * submission will never recurse a channels submit routine.
340 * There are checks in async_tx.c to prevent this.
342 spin_lock_nested(&iop_chan
->lock
, SINGLE_DEPTH_NESTING
);
343 __iop_adma_slot_cleanup(iop_chan
);
344 spin_unlock(&iop_chan
->lock
);
347 static struct iop_adma_desc_slot
*
348 iop_adma_alloc_slots(struct iop_adma_chan
*iop_chan
, int num_slots
,
351 struct iop_adma_desc_slot
*iter
, *_iter
, *alloc_start
= NULL
;
353 int slots_found
, retry
= 0;
355 /* start search from the last allocated descrtiptor
356 * if a contiguous allocation can not be found start searching
357 * from the beginning of the list
362 iter
= iop_chan
->last_used
;
364 iter
= list_entry(&iop_chan
->all_slots
,
365 struct iop_adma_desc_slot
,
368 list_for_each_entry_safe_continue(
369 iter
, _iter
, &iop_chan
->all_slots
, slot_node
) {
371 prefetch(&_iter
->async_tx
);
372 if (iter
->slots_per_op
) {
373 /* give up after finding the first busy slot
374 * on the second pass through the list
383 /* start the allocation if the slot is correctly aligned */
384 if (!slots_found
++) {
385 if (iop_desc_is_aligned(iter
, slots_per_op
))
393 if (slots_found
== num_slots
) {
394 struct iop_adma_desc_slot
*alloc_tail
= NULL
;
395 struct iop_adma_desc_slot
*last_used
= NULL
;
399 dev_dbg(iop_chan
->device
->common
.dev
,
400 "allocated slot: %d "
401 "(desc %p phys: %#x) slots_per_op %d\n",
402 iter
->idx
, iter
->hw_desc
,
403 iter
->async_tx
.phys
, slots_per_op
);
405 /* pre-ack all but the last descriptor */
406 if (num_slots
!= slots_per_op
)
407 async_tx_ack(&iter
->async_tx
);
409 list_add_tail(&iter
->chain_node
, &chain
);
411 iter
->async_tx
.cookie
= 0;
412 iter
->slot_cnt
= num_slots
;
413 iter
->xor_check_result
= NULL
;
414 for (i
= 0; i
< slots_per_op
; i
++) {
415 iter
->slots_per_op
= slots_per_op
- i
;
417 iter
= list_entry(iter
->slot_node
.next
,
418 struct iop_adma_desc_slot
,
421 num_slots
-= slots_per_op
;
423 alloc_tail
->group_head
= alloc_start
;
424 alloc_tail
->async_tx
.cookie
= -EBUSY
;
425 list_splice(&chain
, &alloc_tail
->tx_list
);
426 iop_chan
->last_used
= last_used
;
427 iop_desc_clear_next_desc(alloc_start
);
428 iop_desc_clear_next_desc(alloc_tail
);
435 /* perform direct reclaim if the allocation fails */
436 __iop_adma_slot_cleanup(iop_chan
);
442 iop_desc_assign_cookie(struct iop_adma_chan
*iop_chan
,
443 struct iop_adma_desc_slot
*desc
)
445 dma_cookie_t cookie
= iop_chan
->common
.cookie
;
449 iop_chan
->common
.cookie
= desc
->async_tx
.cookie
= cookie
;
453 static void iop_adma_check_threshold(struct iop_adma_chan
*iop_chan
)
455 dev_dbg(iop_chan
->device
->common
.dev
, "pending: %d\n",
458 if (iop_chan
->pending
>= IOP_ADMA_THRESHOLD
) {
459 iop_chan
->pending
= 0;
460 iop_chan_append(iop_chan
);
465 iop_adma_tx_submit(struct dma_async_tx_descriptor
*tx
)
467 struct iop_adma_desc_slot
*sw_desc
= tx_to_iop_adma_slot(tx
);
468 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(tx
->chan
);
469 struct iop_adma_desc_slot
*grp_start
, *old_chain_tail
;
475 grp_start
= sw_desc
->group_head
;
476 slot_cnt
= grp_start
->slot_cnt
;
477 slots_per_op
= grp_start
->slots_per_op
;
479 spin_lock_bh(&iop_chan
->lock
);
480 cookie
= iop_desc_assign_cookie(iop_chan
, sw_desc
);
482 old_chain_tail
= list_entry(iop_chan
->chain
.prev
,
483 struct iop_adma_desc_slot
, chain_node
);
484 list_splice_init(&sw_desc
->tx_list
,
485 &old_chain_tail
->chain_node
);
487 /* fix up the hardware chain */
488 next_dma
= grp_start
->async_tx
.phys
;
489 iop_desc_set_next_desc(old_chain_tail
, next_dma
);
490 BUG_ON(iop_desc_get_next_desc(old_chain_tail
) != next_dma
); /* flush */
492 /* check for pre-chained descriptors */
493 iop_paranoia(iop_desc_get_next_desc(sw_desc
));
495 /* increment the pending count by the number of slots
496 * memcpy operations have a 1:1 (slot:operation) relation
497 * other operations are heavier and will pop the threshold
500 iop_chan
->pending
+= slot_cnt
;
501 iop_adma_check_threshold(iop_chan
);
502 spin_unlock_bh(&iop_chan
->lock
);
504 dev_dbg(iop_chan
->device
->common
.dev
, "%s cookie: %d slot: %d\n",
505 __func__
, sw_desc
->async_tx
.cookie
, sw_desc
->idx
);
510 static void iop_chan_start_null_memcpy(struct iop_adma_chan
*iop_chan
);
511 static void iop_chan_start_null_xor(struct iop_adma_chan
*iop_chan
);
514 * iop_adma_alloc_chan_resources - returns the number of allocated descriptors
515 * @chan - allocate descriptor resources for this channel
516 * @client - current client requesting the channel be ready for requests
518 * Note: We keep the slots for 1 operation on iop_chan->chain at all times. To
519 * avoid deadlock, via async_xor, num_descs_in_pool must at a minimum be
520 * greater than 2x the number slots needed to satisfy a device->max_xor
523 static int iop_adma_alloc_chan_resources(struct dma_chan
*chan
)
527 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
528 struct iop_adma_desc_slot
*slot
= NULL
;
529 int init
= iop_chan
->slots_allocated
? 0 : 1;
530 struct iop_adma_platform_data
*plat_data
=
531 iop_chan
->device
->pdev
->dev
.platform_data
;
532 int num_descs_in_pool
= plat_data
->pool_size
/IOP_ADMA_SLOT_SIZE
;
534 /* Allocate descriptor slots */
536 idx
= iop_chan
->slots_allocated
;
537 if (idx
== num_descs_in_pool
)
540 slot
= kzalloc(sizeof(*slot
), GFP_KERNEL
);
542 printk(KERN_INFO
"IOP ADMA Channel only initialized"
543 " %d descriptor slots", idx
);
546 hw_desc
= (char *) iop_chan
->device
->dma_desc_pool_virt
;
547 slot
->hw_desc
= (void *) &hw_desc
[idx
* IOP_ADMA_SLOT_SIZE
];
549 dma_async_tx_descriptor_init(&slot
->async_tx
, chan
);
550 slot
->async_tx
.tx_submit
= iop_adma_tx_submit
;
551 INIT_LIST_HEAD(&slot
->tx_list
);
552 INIT_LIST_HEAD(&slot
->chain_node
);
553 INIT_LIST_HEAD(&slot
->slot_node
);
554 hw_desc
= (char *) iop_chan
->device
->dma_desc_pool
;
555 slot
->async_tx
.phys
=
556 (dma_addr_t
) &hw_desc
[idx
* IOP_ADMA_SLOT_SIZE
];
559 spin_lock_bh(&iop_chan
->lock
);
560 iop_chan
->slots_allocated
++;
561 list_add_tail(&slot
->slot_node
, &iop_chan
->all_slots
);
562 spin_unlock_bh(&iop_chan
->lock
);
563 } while (iop_chan
->slots_allocated
< num_descs_in_pool
);
565 if (idx
&& !iop_chan
->last_used
)
566 iop_chan
->last_used
= list_entry(iop_chan
->all_slots
.next
,
567 struct iop_adma_desc_slot
,
570 dev_dbg(iop_chan
->device
->common
.dev
,
571 "allocated %d descriptor slots last_used: %p\n",
572 iop_chan
->slots_allocated
, iop_chan
->last_used
);
574 /* initialize the channel and the chain with a null operation */
576 if (dma_has_cap(DMA_MEMCPY
,
577 iop_chan
->device
->common
.cap_mask
))
578 iop_chan_start_null_memcpy(iop_chan
);
579 else if (dma_has_cap(DMA_XOR
,
580 iop_chan
->device
->common
.cap_mask
))
581 iop_chan_start_null_xor(iop_chan
);
586 return (idx
> 0) ? idx
: -ENOMEM
;
589 static struct dma_async_tx_descriptor
*
590 iop_adma_prep_dma_interrupt(struct dma_chan
*chan
, unsigned long flags
)
592 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
593 struct iop_adma_desc_slot
*sw_desc
, *grp_start
;
594 int slot_cnt
, slots_per_op
;
596 dev_dbg(iop_chan
->device
->common
.dev
, "%s\n", __func__
);
598 spin_lock_bh(&iop_chan
->lock
);
599 slot_cnt
= iop_chan_interrupt_slot_count(&slots_per_op
, iop_chan
);
600 sw_desc
= iop_adma_alloc_slots(iop_chan
, slot_cnt
, slots_per_op
);
602 grp_start
= sw_desc
->group_head
;
603 iop_desc_init_interrupt(grp_start
, iop_chan
);
604 grp_start
->unmap_len
= 0;
605 sw_desc
->async_tx
.flags
= flags
;
607 spin_unlock_bh(&iop_chan
->lock
);
609 return sw_desc
? &sw_desc
->async_tx
: NULL
;
612 static struct dma_async_tx_descriptor
*
613 iop_adma_prep_dma_memcpy(struct dma_chan
*chan
, dma_addr_t dma_dest
,
614 dma_addr_t dma_src
, size_t len
, unsigned long flags
)
616 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
617 struct iop_adma_desc_slot
*sw_desc
, *grp_start
;
618 int slot_cnt
, slots_per_op
;
622 BUG_ON(unlikely(len
> IOP_ADMA_MAX_BYTE_COUNT
));
624 dev_dbg(iop_chan
->device
->common
.dev
, "%s len: %u\n",
627 spin_lock_bh(&iop_chan
->lock
);
628 slot_cnt
= iop_chan_memcpy_slot_count(len
, &slots_per_op
);
629 sw_desc
= iop_adma_alloc_slots(iop_chan
, slot_cnt
, slots_per_op
);
631 grp_start
= sw_desc
->group_head
;
632 iop_desc_init_memcpy(grp_start
, flags
);
633 iop_desc_set_byte_count(grp_start
, iop_chan
, len
);
634 iop_desc_set_dest_addr(grp_start
, iop_chan
, dma_dest
);
635 iop_desc_set_memcpy_src_addr(grp_start
, dma_src
);
636 sw_desc
->unmap_src_cnt
= 1;
637 sw_desc
->unmap_len
= len
;
638 sw_desc
->async_tx
.flags
= flags
;
640 spin_unlock_bh(&iop_chan
->lock
);
642 return sw_desc
? &sw_desc
->async_tx
: NULL
;
645 static struct dma_async_tx_descriptor
*
646 iop_adma_prep_dma_memset(struct dma_chan
*chan
, dma_addr_t dma_dest
,
647 int value
, size_t len
, unsigned long flags
)
649 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
650 struct iop_adma_desc_slot
*sw_desc
, *grp_start
;
651 int slot_cnt
, slots_per_op
;
655 BUG_ON(unlikely(len
> IOP_ADMA_MAX_BYTE_COUNT
));
657 dev_dbg(iop_chan
->device
->common
.dev
, "%s len: %u\n",
660 spin_lock_bh(&iop_chan
->lock
);
661 slot_cnt
= iop_chan_memset_slot_count(len
, &slots_per_op
);
662 sw_desc
= iop_adma_alloc_slots(iop_chan
, slot_cnt
, slots_per_op
);
664 grp_start
= sw_desc
->group_head
;
665 iop_desc_init_memset(grp_start
, flags
);
666 iop_desc_set_byte_count(grp_start
, iop_chan
, len
);
667 iop_desc_set_block_fill_val(grp_start
, value
);
668 iop_desc_set_dest_addr(grp_start
, iop_chan
, dma_dest
);
669 sw_desc
->unmap_src_cnt
= 1;
670 sw_desc
->unmap_len
= len
;
671 sw_desc
->async_tx
.flags
= flags
;
673 spin_unlock_bh(&iop_chan
->lock
);
675 return sw_desc
? &sw_desc
->async_tx
: NULL
;
678 static struct dma_async_tx_descriptor
*
679 iop_adma_prep_dma_xor(struct dma_chan
*chan
, dma_addr_t dma_dest
,
680 dma_addr_t
*dma_src
, unsigned int src_cnt
, size_t len
,
683 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
684 struct iop_adma_desc_slot
*sw_desc
, *grp_start
;
685 int slot_cnt
, slots_per_op
;
689 BUG_ON(unlikely(len
> IOP_ADMA_XOR_MAX_BYTE_COUNT
));
691 dev_dbg(iop_chan
->device
->common
.dev
,
692 "%s src_cnt: %d len: %u flags: %lx\n",
693 __func__
, src_cnt
, len
, flags
);
695 spin_lock_bh(&iop_chan
->lock
);
696 slot_cnt
= iop_chan_xor_slot_count(len
, src_cnt
, &slots_per_op
);
697 sw_desc
= iop_adma_alloc_slots(iop_chan
, slot_cnt
, slots_per_op
);
699 grp_start
= sw_desc
->group_head
;
700 iop_desc_init_xor(grp_start
, src_cnt
, flags
);
701 iop_desc_set_byte_count(grp_start
, iop_chan
, len
);
702 iop_desc_set_dest_addr(grp_start
, iop_chan
, dma_dest
);
703 sw_desc
->unmap_src_cnt
= src_cnt
;
704 sw_desc
->unmap_len
= len
;
705 sw_desc
->async_tx
.flags
= flags
;
707 iop_desc_set_xor_src_addr(grp_start
, src_cnt
,
710 spin_unlock_bh(&iop_chan
->lock
);
712 return sw_desc
? &sw_desc
->async_tx
: NULL
;
715 static struct dma_async_tx_descriptor
*
716 iop_adma_prep_dma_xor_val(struct dma_chan
*chan
, dma_addr_t
*dma_src
,
717 unsigned int src_cnt
, size_t len
, u32
*result
,
720 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
721 struct iop_adma_desc_slot
*sw_desc
, *grp_start
;
722 int slot_cnt
, slots_per_op
;
727 dev_dbg(iop_chan
->device
->common
.dev
, "%s src_cnt: %d len: %u\n",
728 __func__
, src_cnt
, len
);
730 spin_lock_bh(&iop_chan
->lock
);
731 slot_cnt
= iop_chan_zero_sum_slot_count(len
, src_cnt
, &slots_per_op
);
732 sw_desc
= iop_adma_alloc_slots(iop_chan
, slot_cnt
, slots_per_op
);
734 grp_start
= sw_desc
->group_head
;
735 iop_desc_init_zero_sum(grp_start
, src_cnt
, flags
);
736 iop_desc_set_zero_sum_byte_count(grp_start
, len
);
737 grp_start
->xor_check_result
= result
;
738 pr_debug("\t%s: grp_start->xor_check_result: %p\n",
739 __func__
, grp_start
->xor_check_result
);
740 sw_desc
->unmap_src_cnt
= src_cnt
;
741 sw_desc
->unmap_len
= len
;
742 sw_desc
->async_tx
.flags
= flags
;
744 iop_desc_set_zero_sum_src_addr(grp_start
, src_cnt
,
747 spin_unlock_bh(&iop_chan
->lock
);
749 return sw_desc
? &sw_desc
->async_tx
: NULL
;
752 static struct dma_async_tx_descriptor
*
753 iop_adma_prep_dma_pq(struct dma_chan
*chan
, dma_addr_t
*dst
, dma_addr_t
*src
,
754 unsigned int src_cnt
, const unsigned char *scf
, size_t len
,
757 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
758 struct iop_adma_desc_slot
*sw_desc
, *g
;
759 int slot_cnt
, slots_per_op
;
764 BUG_ON(len
> IOP_ADMA_XOR_MAX_BYTE_COUNT
);
766 dev_dbg(iop_chan
->device
->common
.dev
,
767 "%s src_cnt: %d len: %u flags: %lx\n",
768 __func__
, src_cnt
, len
, flags
);
770 if (dmaf_p_disabled_continue(flags
))
771 continue_srcs
= 1+src_cnt
;
772 else if (dmaf_continue(flags
))
773 continue_srcs
= 3+src_cnt
;
775 continue_srcs
= 0+src_cnt
;
777 spin_lock_bh(&iop_chan
->lock
);
778 slot_cnt
= iop_chan_pq_slot_count(len
, continue_srcs
, &slots_per_op
);
779 sw_desc
= iop_adma_alloc_slots(iop_chan
, slot_cnt
, slots_per_op
);
783 g
= sw_desc
->group_head
;
784 iop_desc_set_byte_count(g
, iop_chan
, len
);
786 /* even if P is disabled its destination address (bits
787 * [3:0]) must match Q. It is ok if P points to an
788 * invalid address, it won't be written.
790 if (flags
& DMA_PREP_PQ_DISABLE_P
)
791 dst
[0] = dst
[1] & 0x7;
793 iop_desc_set_pq_addr(g
, dst
);
794 sw_desc
->unmap_src_cnt
= src_cnt
;
795 sw_desc
->unmap_len
= len
;
796 sw_desc
->async_tx
.flags
= flags
;
797 for (i
= 0; i
< src_cnt
; i
++)
798 iop_desc_set_pq_src_addr(g
, i
, src
[i
], scf
[i
]);
800 /* if we are continuing a previous operation factor in
801 * the old p and q values, see the comment for dma_maxpq
802 * in include/linux/dmaengine.h
804 if (dmaf_p_disabled_continue(flags
))
805 iop_desc_set_pq_src_addr(g
, i
++, dst
[1], 1);
806 else if (dmaf_continue(flags
)) {
807 iop_desc_set_pq_src_addr(g
, i
++, dst
[0], 0);
808 iop_desc_set_pq_src_addr(g
, i
++, dst
[1], 1);
809 iop_desc_set_pq_src_addr(g
, i
++, dst
[1], 0);
811 iop_desc_init_pq(g
, i
, flags
);
813 spin_unlock_bh(&iop_chan
->lock
);
815 return sw_desc
? &sw_desc
->async_tx
: NULL
;
818 static struct dma_async_tx_descriptor
*
819 iop_adma_prep_dma_pq_val(struct dma_chan
*chan
, dma_addr_t
*pq
, dma_addr_t
*src
,
820 unsigned int src_cnt
, const unsigned char *scf
,
821 size_t len
, enum sum_check_flags
*pqres
,
824 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
825 struct iop_adma_desc_slot
*sw_desc
, *g
;
826 int slot_cnt
, slots_per_op
;
830 BUG_ON(len
> IOP_ADMA_XOR_MAX_BYTE_COUNT
);
832 dev_dbg(iop_chan
->device
->common
.dev
, "%s src_cnt: %d len: %u\n",
833 __func__
, src_cnt
, len
);
835 spin_lock_bh(&iop_chan
->lock
);
836 slot_cnt
= iop_chan_pq_zero_sum_slot_count(len
, src_cnt
+ 2, &slots_per_op
);
837 sw_desc
= iop_adma_alloc_slots(iop_chan
, slot_cnt
, slots_per_op
);
839 /* for validate operations p and q are tagged onto the
840 * end of the source list
842 int pq_idx
= src_cnt
;
844 g
= sw_desc
->group_head
;
845 iop_desc_init_pq_zero_sum(g
, src_cnt
+2, flags
);
846 iop_desc_set_pq_zero_sum_byte_count(g
, len
);
847 g
->pq_check_result
= pqres
;
848 pr_debug("\t%s: g->pq_check_result: %p\n",
849 __func__
, g
->pq_check_result
);
850 sw_desc
->unmap_src_cnt
= src_cnt
+2;
851 sw_desc
->unmap_len
= len
;
852 sw_desc
->async_tx
.flags
= flags
;
854 iop_desc_set_pq_zero_sum_src_addr(g
, src_cnt
,
857 iop_desc_set_pq_zero_sum_addr(g
, pq_idx
, src
);
859 spin_unlock_bh(&iop_chan
->lock
);
861 return sw_desc
? &sw_desc
->async_tx
: NULL
;
864 static void iop_adma_free_chan_resources(struct dma_chan
*chan
)
866 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
867 struct iop_adma_desc_slot
*iter
, *_iter
;
868 int in_use_descs
= 0;
870 iop_adma_slot_cleanup(iop_chan
);
872 spin_lock_bh(&iop_chan
->lock
);
873 list_for_each_entry_safe(iter
, _iter
, &iop_chan
->chain
,
876 list_del(&iter
->chain_node
);
878 list_for_each_entry_safe_reverse(
879 iter
, _iter
, &iop_chan
->all_slots
, slot_node
) {
880 list_del(&iter
->slot_node
);
882 iop_chan
->slots_allocated
--;
884 iop_chan
->last_used
= NULL
;
886 dev_dbg(iop_chan
->device
->common
.dev
, "%s slots_allocated %d\n",
887 __func__
, iop_chan
->slots_allocated
);
888 spin_unlock_bh(&iop_chan
->lock
);
890 /* one is ok since we left it on there on purpose */
891 if (in_use_descs
> 1)
892 printk(KERN_ERR
"IOP: Freeing %d in use descriptors!\n",
897 * iop_adma_is_complete - poll the status of an ADMA transaction
898 * @chan: ADMA channel handle
899 * @cookie: ADMA transaction identifier
901 static enum dma_status
iop_adma_is_complete(struct dma_chan
*chan
,
906 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
907 dma_cookie_t last_used
;
908 dma_cookie_t last_complete
;
911 last_used
= chan
->cookie
;
912 last_complete
= iop_chan
->completed_cookie
;
915 *done
= last_complete
;
919 ret
= dma_async_is_complete(cookie
, last_complete
, last_used
);
920 if (ret
== DMA_SUCCESS
)
923 iop_adma_slot_cleanup(iop_chan
);
925 last_used
= chan
->cookie
;
926 last_complete
= iop_chan
->completed_cookie
;
929 *done
= last_complete
;
933 return dma_async_is_complete(cookie
, last_complete
, last_used
);
936 static irqreturn_t
iop_adma_eot_handler(int irq
, void *data
)
938 struct iop_adma_chan
*chan
= data
;
940 dev_dbg(chan
->device
->common
.dev
, "%s\n", __func__
);
942 tasklet_schedule(&chan
->irq_tasklet
);
944 iop_adma_device_clear_eot_status(chan
);
949 static irqreturn_t
iop_adma_eoc_handler(int irq
, void *data
)
951 struct iop_adma_chan
*chan
= data
;
953 dev_dbg(chan
->device
->common
.dev
, "%s\n", __func__
);
955 tasklet_schedule(&chan
->irq_tasklet
);
957 iop_adma_device_clear_eoc_status(chan
);
962 static irqreturn_t
iop_adma_err_handler(int irq
, void *data
)
964 struct iop_adma_chan
*chan
= data
;
965 unsigned long status
= iop_chan_get_status(chan
);
967 dev_printk(KERN_ERR
, chan
->device
->common
.dev
,
968 "error ( %s%s%s%s%s%s%s)\n",
969 iop_is_err_int_parity(status
, chan
) ? "int_parity " : "",
970 iop_is_err_mcu_abort(status
, chan
) ? "mcu_abort " : "",
971 iop_is_err_int_tabort(status
, chan
) ? "int_tabort " : "",
972 iop_is_err_int_mabort(status
, chan
) ? "int_mabort " : "",
973 iop_is_err_pci_tabort(status
, chan
) ? "pci_tabort " : "",
974 iop_is_err_pci_mabort(status
, chan
) ? "pci_mabort " : "",
975 iop_is_err_split_tx(status
, chan
) ? "split_tx " : "");
977 iop_adma_device_clear_err_status(chan
);
984 static void iop_adma_issue_pending(struct dma_chan
*chan
)
986 struct iop_adma_chan
*iop_chan
= to_iop_adma_chan(chan
);
988 if (iop_chan
->pending
) {
989 iop_chan
->pending
= 0;
990 iop_chan_append(iop_chan
);
995 * Perform a transaction to verify the HW works.
997 #define IOP_ADMA_TEST_SIZE 2000
999 static int __devinit
iop_adma_memcpy_self_test(struct iop_adma_device
*device
)
1003 dma_addr_t src_dma
, dest_dma
;
1004 struct dma_chan
*dma_chan
;
1005 dma_cookie_t cookie
;
1006 struct dma_async_tx_descriptor
*tx
;
1008 struct iop_adma_chan
*iop_chan
;
1010 dev_dbg(device
->common
.dev
, "%s\n", __func__
);
1012 src
= kmalloc(IOP_ADMA_TEST_SIZE
, GFP_KERNEL
);
1015 dest
= kzalloc(IOP_ADMA_TEST_SIZE
, GFP_KERNEL
);
1021 /* Fill in src buffer */
1022 for (i
= 0; i
< IOP_ADMA_TEST_SIZE
; i
++)
1023 ((u8
*) src
)[i
] = (u8
)i
;
1025 /* Start copy, using first DMA channel */
1026 dma_chan
= container_of(device
->common
.channels
.next
,
1029 if (iop_adma_alloc_chan_resources(dma_chan
) < 1) {
1034 dest_dma
= dma_map_single(dma_chan
->device
->dev
, dest
,
1035 IOP_ADMA_TEST_SIZE
, DMA_FROM_DEVICE
);
1036 src_dma
= dma_map_single(dma_chan
->device
->dev
, src
,
1037 IOP_ADMA_TEST_SIZE
, DMA_TO_DEVICE
);
1038 tx
= iop_adma_prep_dma_memcpy(dma_chan
, dest_dma
, src_dma
,
1040 DMA_PREP_INTERRUPT
| DMA_CTRL_ACK
);
1042 cookie
= iop_adma_tx_submit(tx
);
1043 iop_adma_issue_pending(dma_chan
);
1046 if (iop_adma_is_complete(dma_chan
, cookie
, NULL
, NULL
) !=
1048 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1049 "Self-test copy timed out, disabling\n");
1051 goto free_resources
;
1054 iop_chan
= to_iop_adma_chan(dma_chan
);
1055 dma_sync_single_for_cpu(&iop_chan
->device
->pdev
->dev
, dest_dma
,
1056 IOP_ADMA_TEST_SIZE
, DMA_FROM_DEVICE
);
1057 if (memcmp(src
, dest
, IOP_ADMA_TEST_SIZE
)) {
1058 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1059 "Self-test copy failed compare, disabling\n");
1061 goto free_resources
;
1065 iop_adma_free_chan_resources(dma_chan
);
1072 #define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
1073 static int __devinit
1074 iop_adma_xor_val_self_test(struct iop_adma_device
*device
)
1078 struct page
*xor_srcs
[IOP_ADMA_NUM_SRC_TEST
];
1079 struct page
*zero_sum_srcs
[IOP_ADMA_NUM_SRC_TEST
+ 1];
1080 dma_addr_t dma_srcs
[IOP_ADMA_NUM_SRC_TEST
+ 1];
1081 dma_addr_t dma_addr
, dest_dma
;
1082 struct dma_async_tx_descriptor
*tx
;
1083 struct dma_chan
*dma_chan
;
1084 dma_cookie_t cookie
;
1087 u32 zero_sum_result
;
1089 struct iop_adma_chan
*iop_chan
;
1091 dev_dbg(device
->common
.dev
, "%s\n", __func__
);
1093 for (src_idx
= 0; src_idx
< IOP_ADMA_NUM_SRC_TEST
; src_idx
++) {
1094 xor_srcs
[src_idx
] = alloc_page(GFP_KERNEL
);
1095 if (!xor_srcs
[src_idx
]) {
1097 __free_page(xor_srcs
[src_idx
]);
1102 dest
= alloc_page(GFP_KERNEL
);
1105 __free_page(xor_srcs
[src_idx
]);
1109 /* Fill in src buffers */
1110 for (src_idx
= 0; src_idx
< IOP_ADMA_NUM_SRC_TEST
; src_idx
++) {
1111 u8
*ptr
= page_address(xor_srcs
[src_idx
]);
1112 for (i
= 0; i
< PAGE_SIZE
; i
++)
1113 ptr
[i
] = (1 << src_idx
);
1116 for (src_idx
= 0; src_idx
< IOP_ADMA_NUM_SRC_TEST
; src_idx
++)
1117 cmp_byte
^= (u8
) (1 << src_idx
);
1119 cmp_word
= (cmp_byte
<< 24) | (cmp_byte
<< 16) |
1120 (cmp_byte
<< 8) | cmp_byte
;
1122 memset(page_address(dest
), 0, PAGE_SIZE
);
1124 dma_chan
= container_of(device
->common
.channels
.next
,
1127 if (iop_adma_alloc_chan_resources(dma_chan
) < 1) {
1133 dest_dma
= dma_map_page(dma_chan
->device
->dev
, dest
, 0,
1134 PAGE_SIZE
, DMA_FROM_DEVICE
);
1135 for (i
= 0; i
< IOP_ADMA_NUM_SRC_TEST
; i
++)
1136 dma_srcs
[i
] = dma_map_page(dma_chan
->device
->dev
, xor_srcs
[i
],
1137 0, PAGE_SIZE
, DMA_TO_DEVICE
);
1138 tx
= iop_adma_prep_dma_xor(dma_chan
, dest_dma
, dma_srcs
,
1139 IOP_ADMA_NUM_SRC_TEST
, PAGE_SIZE
,
1140 DMA_PREP_INTERRUPT
| DMA_CTRL_ACK
);
1142 cookie
= iop_adma_tx_submit(tx
);
1143 iop_adma_issue_pending(dma_chan
);
1146 if (iop_adma_is_complete(dma_chan
, cookie
, NULL
, NULL
) !=
1148 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1149 "Self-test xor timed out, disabling\n");
1151 goto free_resources
;
1154 iop_chan
= to_iop_adma_chan(dma_chan
);
1155 dma_sync_single_for_cpu(&iop_chan
->device
->pdev
->dev
, dest_dma
,
1156 PAGE_SIZE
, DMA_FROM_DEVICE
);
1157 for (i
= 0; i
< (PAGE_SIZE
/ sizeof(u32
)); i
++) {
1158 u32
*ptr
= page_address(dest
);
1159 if (ptr
[i
] != cmp_word
) {
1160 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1161 "Self-test xor failed compare, disabling\n");
1163 goto free_resources
;
1166 dma_sync_single_for_device(&iop_chan
->device
->pdev
->dev
, dest_dma
,
1167 PAGE_SIZE
, DMA_TO_DEVICE
);
1169 /* skip zero sum if the capability is not present */
1170 if (!dma_has_cap(DMA_XOR_VAL
, dma_chan
->device
->cap_mask
))
1171 goto free_resources
;
1173 /* zero sum the sources with the destintation page */
1174 for (i
= 0; i
< IOP_ADMA_NUM_SRC_TEST
; i
++)
1175 zero_sum_srcs
[i
] = xor_srcs
[i
];
1176 zero_sum_srcs
[i
] = dest
;
1178 zero_sum_result
= 1;
1180 for (i
= 0; i
< IOP_ADMA_NUM_SRC_TEST
+ 1; i
++)
1181 dma_srcs
[i
] = dma_map_page(dma_chan
->device
->dev
,
1182 zero_sum_srcs
[i
], 0, PAGE_SIZE
,
1184 tx
= iop_adma_prep_dma_xor_val(dma_chan
, dma_srcs
,
1185 IOP_ADMA_NUM_SRC_TEST
+ 1, PAGE_SIZE
,
1187 DMA_PREP_INTERRUPT
| DMA_CTRL_ACK
);
1189 cookie
= iop_adma_tx_submit(tx
);
1190 iop_adma_issue_pending(dma_chan
);
1193 if (iop_adma_is_complete(dma_chan
, cookie
, NULL
, NULL
) != DMA_SUCCESS
) {
1194 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1195 "Self-test zero sum timed out, disabling\n");
1197 goto free_resources
;
1200 if (zero_sum_result
!= 0) {
1201 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1202 "Self-test zero sum failed compare, disabling\n");
1204 goto free_resources
;
1208 dma_addr
= dma_map_page(dma_chan
->device
->dev
, dest
, 0,
1209 PAGE_SIZE
, DMA_FROM_DEVICE
);
1210 tx
= iop_adma_prep_dma_memset(dma_chan
, dma_addr
, 0, PAGE_SIZE
,
1211 DMA_PREP_INTERRUPT
| DMA_CTRL_ACK
);
1213 cookie
= iop_adma_tx_submit(tx
);
1214 iop_adma_issue_pending(dma_chan
);
1217 if (iop_adma_is_complete(dma_chan
, cookie
, NULL
, NULL
) != DMA_SUCCESS
) {
1218 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1219 "Self-test memset timed out, disabling\n");
1221 goto free_resources
;
1224 for (i
= 0; i
< PAGE_SIZE
/sizeof(u32
); i
++) {
1225 u32
*ptr
= page_address(dest
);
1227 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1228 "Self-test memset failed compare, disabling\n");
1230 goto free_resources
;
1234 /* test for non-zero parity sum */
1235 zero_sum_result
= 0;
1236 for (i
= 0; i
< IOP_ADMA_NUM_SRC_TEST
+ 1; i
++)
1237 dma_srcs
[i
] = dma_map_page(dma_chan
->device
->dev
,
1238 zero_sum_srcs
[i
], 0, PAGE_SIZE
,
1240 tx
= iop_adma_prep_dma_xor_val(dma_chan
, dma_srcs
,
1241 IOP_ADMA_NUM_SRC_TEST
+ 1, PAGE_SIZE
,
1243 DMA_PREP_INTERRUPT
| DMA_CTRL_ACK
);
1245 cookie
= iop_adma_tx_submit(tx
);
1246 iop_adma_issue_pending(dma_chan
);
1249 if (iop_adma_is_complete(dma_chan
, cookie
, NULL
, NULL
) != DMA_SUCCESS
) {
1250 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1251 "Self-test non-zero sum timed out, disabling\n");
1253 goto free_resources
;
1256 if (zero_sum_result
!= 1) {
1257 dev_printk(KERN_ERR
, dma_chan
->device
->dev
,
1258 "Self-test non-zero sum failed compare, disabling\n");
1260 goto free_resources
;
1264 iop_adma_free_chan_resources(dma_chan
);
1266 src_idx
= IOP_ADMA_NUM_SRC_TEST
;
1268 __free_page(xor_srcs
[src_idx
]);
1273 #ifdef CONFIG_MD_RAID6_PQ
1274 static int __devinit
1275 iop_adma_pq_zero_sum_self_test(struct iop_adma_device
*device
)
1277 /* combined sources, software pq results, and extra hw pq results */
1278 struct page
*pq
[IOP_ADMA_NUM_SRC_TEST
+2+2];
1279 /* ptr to the extra hw pq buffers defined above */
1280 struct page
**pq_hw
= &pq
[IOP_ADMA_NUM_SRC_TEST
+2];
1281 /* address conversion buffers (dma_map / page_address) */
1282 void *pq_sw
[IOP_ADMA_NUM_SRC_TEST
+2];
1283 dma_addr_t pq_src
[IOP_ADMA_NUM_SRC_TEST
];
1284 dma_addr_t pq_dest
[2];
1287 struct dma_async_tx_descriptor
*tx
;
1288 struct dma_chan
*dma_chan
;
1289 dma_cookie_t cookie
;
1290 u32 zero_sum_result
;
1294 dev_dbg(device
->common
.dev
, "%s\n", __func__
);
1296 for (i
= 0; i
< ARRAY_SIZE(pq
); i
++) {
1297 pq
[i
] = alloc_page(GFP_KERNEL
);
1305 /* Fill in src buffers */
1306 for (i
= 0; i
< IOP_ADMA_NUM_SRC_TEST
; i
++) {
1307 pq_sw
[i
] = page_address(pq
[i
]);
1308 memset(pq_sw
[i
], 0x11111111 * (1<<i
), PAGE_SIZE
);
1310 pq_sw
[i
] = page_address(pq
[i
]);
1311 pq_sw
[i
+1] = page_address(pq
[i
+1]);
1313 dma_chan
= container_of(device
->common
.channels
.next
,
1316 if (iop_adma_alloc_chan_resources(dma_chan
) < 1) {
1321 dev
= dma_chan
->device
->dev
;
1323 /* initialize the dests */
1324 memset(page_address(pq_hw
[0]), 0 , PAGE_SIZE
);
1325 memset(page_address(pq_hw
[1]), 0 , PAGE_SIZE
);
1328 pq_dest
[0] = dma_map_page(dev
, pq_hw
[0], 0, PAGE_SIZE
, DMA_FROM_DEVICE
);
1329 pq_dest
[1] = dma_map_page(dev
, pq_hw
[1], 0, PAGE_SIZE
, DMA_FROM_DEVICE
);
1330 for (i
= 0; i
< IOP_ADMA_NUM_SRC_TEST
; i
++)
1331 pq_src
[i
] = dma_map_page(dev
, pq
[i
], 0, PAGE_SIZE
,
1334 tx
= iop_adma_prep_dma_pq(dma_chan
, pq_dest
, pq_src
,
1335 IOP_ADMA_NUM_SRC_TEST
, (u8
*)raid6_gfexp
,
1337 DMA_PREP_INTERRUPT
|
1340 cookie
= iop_adma_tx_submit(tx
);
1341 iop_adma_issue_pending(dma_chan
);
1344 if (iop_adma_is_complete(dma_chan
, cookie
, NULL
, NULL
) !=
1346 dev_err(dev
, "Self-test pq timed out, disabling\n");
1348 goto free_resources
;
1351 raid6_call
.gen_syndrome(IOP_ADMA_NUM_SRC_TEST
+2, PAGE_SIZE
, pq_sw
);
1353 if (memcmp(pq_sw
[IOP_ADMA_NUM_SRC_TEST
],
1354 page_address(pq_hw
[0]), PAGE_SIZE
) != 0) {
1355 dev_err(dev
, "Self-test p failed compare, disabling\n");
1357 goto free_resources
;
1359 if (memcmp(pq_sw
[IOP_ADMA_NUM_SRC_TEST
+1],
1360 page_address(pq_hw
[1]), PAGE_SIZE
) != 0) {
1361 dev_err(dev
, "Self-test q failed compare, disabling\n");
1363 goto free_resources
;
1366 /* test correct zero sum using the software generated pq values */
1367 for (i
= 0; i
< IOP_ADMA_NUM_SRC_TEST
+ 2; i
++)
1368 pq_src
[i
] = dma_map_page(dev
, pq
[i
], 0, PAGE_SIZE
,
1371 zero_sum_result
= ~0;
1372 tx
= iop_adma_prep_dma_pq_val(dma_chan
, &pq_src
[IOP_ADMA_NUM_SRC_TEST
],
1373 pq_src
, IOP_ADMA_NUM_SRC_TEST
,
1374 raid6_gfexp
, PAGE_SIZE
, &zero_sum_result
,
1375 DMA_PREP_INTERRUPT
|DMA_CTRL_ACK
);
1377 cookie
= iop_adma_tx_submit(tx
);
1378 iop_adma_issue_pending(dma_chan
);
1381 if (iop_adma_is_complete(dma_chan
, cookie
, NULL
, NULL
) !=
1383 dev_err(dev
, "Self-test pq-zero-sum timed out, disabling\n");
1385 goto free_resources
;
1388 if (zero_sum_result
!= 0) {
1389 dev_err(dev
, "Self-test pq-zero-sum failed to validate: %x\n",
1392 goto free_resources
;
1395 /* test incorrect zero sum */
1396 i
= IOP_ADMA_NUM_SRC_TEST
;
1397 memset(pq_sw
[i
] + 100, 0, 100);
1398 memset(pq_sw
[i
+1] + 200, 0, 200);
1399 for (i
= 0; i
< IOP_ADMA_NUM_SRC_TEST
+ 2; i
++)
1400 pq_src
[i
] = dma_map_page(dev
, pq
[i
], 0, PAGE_SIZE
,
1403 zero_sum_result
= 0;
1404 tx
= iop_adma_prep_dma_pq_val(dma_chan
, &pq_src
[IOP_ADMA_NUM_SRC_TEST
],
1405 pq_src
, IOP_ADMA_NUM_SRC_TEST
,
1406 raid6_gfexp
, PAGE_SIZE
, &zero_sum_result
,
1407 DMA_PREP_INTERRUPT
|DMA_CTRL_ACK
);
1409 cookie
= iop_adma_tx_submit(tx
);
1410 iop_adma_issue_pending(dma_chan
);
1413 if (iop_adma_is_complete(dma_chan
, cookie
, NULL
, NULL
) !=
1415 dev_err(dev
, "Self-test !pq-zero-sum timed out, disabling\n");
1417 goto free_resources
;
1420 if (zero_sum_result
!= (SUM_CHECK_P_RESULT
| SUM_CHECK_Q_RESULT
)) {
1421 dev_err(dev
, "Self-test !pq-zero-sum failed to validate: %x\n",
1424 goto free_resources
;
1428 iop_adma_free_chan_resources(dma_chan
);
1437 static int __devexit
iop_adma_remove(struct platform_device
*dev
)
1439 struct iop_adma_device
*device
= platform_get_drvdata(dev
);
1440 struct dma_chan
*chan
, *_chan
;
1441 struct iop_adma_chan
*iop_chan
;
1442 struct iop_adma_platform_data
*plat_data
= dev
->dev
.platform_data
;
1444 dma_async_device_unregister(&device
->common
);
1446 dma_free_coherent(&dev
->dev
, plat_data
->pool_size
,
1447 device
->dma_desc_pool_virt
, device
->dma_desc_pool
);
1449 list_for_each_entry_safe(chan
, _chan
, &device
->common
.channels
,
1451 iop_chan
= to_iop_adma_chan(chan
);
1452 list_del(&chan
->device_node
);
1460 static int __devinit
iop_adma_probe(struct platform_device
*pdev
)
1462 struct resource
*res
;
1464 struct iop_adma_device
*adev
;
1465 struct iop_adma_chan
*iop_chan
;
1466 struct dma_device
*dma_dev
;
1467 struct iop_adma_platform_data
*plat_data
= pdev
->dev
.platform_data
;
1469 res
= platform_get_resource(pdev
, IORESOURCE_MEM
, 0);
1473 if (!devm_request_mem_region(&pdev
->dev
, res
->start
,
1474 resource_size(res
), pdev
->name
))
1477 adev
= kzalloc(sizeof(*adev
), GFP_KERNEL
);
1480 dma_dev
= &adev
->common
;
1482 /* allocate coherent memory for hardware descriptors
1483 * note: writecombine gives slightly better performance, but
1484 * requires that we explicitly flush the writes
1486 if ((adev
->dma_desc_pool_virt
= dma_alloc_writecombine(&pdev
->dev
,
1487 plat_data
->pool_size
,
1488 &adev
->dma_desc_pool
,
1489 GFP_KERNEL
)) == NULL
) {
1494 dev_dbg(&pdev
->dev
, "%s: allocted descriptor pool virt %p phys %p\n",
1495 __func__
, adev
->dma_desc_pool_virt
,
1496 (void *) adev
->dma_desc_pool
);
1498 adev
->id
= plat_data
->hw_id
;
1500 /* discover transaction capabilites from the platform data */
1501 dma_dev
->cap_mask
= plat_data
->cap_mask
;
1504 platform_set_drvdata(pdev
, adev
);
1506 INIT_LIST_HEAD(&dma_dev
->channels
);
1508 /* set base routines */
1509 dma_dev
->device_alloc_chan_resources
= iop_adma_alloc_chan_resources
;
1510 dma_dev
->device_free_chan_resources
= iop_adma_free_chan_resources
;
1511 dma_dev
->device_is_tx_complete
= iop_adma_is_complete
;
1512 dma_dev
->device_issue_pending
= iop_adma_issue_pending
;
1513 dma_dev
->dev
= &pdev
->dev
;
1515 /* set prep routines based on capability */
1516 if (dma_has_cap(DMA_MEMCPY
, dma_dev
->cap_mask
))
1517 dma_dev
->device_prep_dma_memcpy
= iop_adma_prep_dma_memcpy
;
1518 if (dma_has_cap(DMA_MEMSET
, dma_dev
->cap_mask
))
1519 dma_dev
->device_prep_dma_memset
= iop_adma_prep_dma_memset
;
1520 if (dma_has_cap(DMA_XOR
, dma_dev
->cap_mask
)) {
1521 dma_dev
->max_xor
= iop_adma_get_max_xor();
1522 dma_dev
->device_prep_dma_xor
= iop_adma_prep_dma_xor
;
1524 if (dma_has_cap(DMA_XOR_VAL
, dma_dev
->cap_mask
))
1525 dma_dev
->device_prep_dma_xor_val
=
1526 iop_adma_prep_dma_xor_val
;
1527 if (dma_has_cap(DMA_PQ
, dma_dev
->cap_mask
)) {
1528 dma_set_maxpq(dma_dev
, iop_adma_get_max_pq(), 0);
1529 dma_dev
->device_prep_dma_pq
= iop_adma_prep_dma_pq
;
1531 if (dma_has_cap(DMA_PQ_VAL
, dma_dev
->cap_mask
))
1532 dma_dev
->device_prep_dma_pq_val
=
1533 iop_adma_prep_dma_pq_val
;
1534 if (dma_has_cap(DMA_INTERRUPT
, dma_dev
->cap_mask
))
1535 dma_dev
->device_prep_dma_interrupt
=
1536 iop_adma_prep_dma_interrupt
;
1538 iop_chan
= kzalloc(sizeof(*iop_chan
), GFP_KERNEL
);
1543 iop_chan
->device
= adev
;
1545 iop_chan
->mmr_base
= devm_ioremap(&pdev
->dev
, res
->start
,
1546 resource_size(res
));
1547 if (!iop_chan
->mmr_base
) {
1549 goto err_free_iop_chan
;
1551 tasklet_init(&iop_chan
->irq_tasklet
, iop_adma_tasklet
, (unsigned long)
1554 /* clear errors before enabling interrupts */
1555 iop_adma_device_clear_err_status(iop_chan
);
1557 for (i
= 0; i
< 3; i
++) {
1558 irq_handler_t handler
[] = { iop_adma_eot_handler
,
1559 iop_adma_eoc_handler
,
1560 iop_adma_err_handler
};
1561 int irq
= platform_get_irq(pdev
, i
);
1564 goto err_free_iop_chan
;
1566 ret
= devm_request_irq(&pdev
->dev
, irq
,
1567 handler
[i
], 0, pdev
->name
, iop_chan
);
1569 goto err_free_iop_chan
;
1573 spin_lock_init(&iop_chan
->lock
);
1574 INIT_LIST_HEAD(&iop_chan
->chain
);
1575 INIT_LIST_HEAD(&iop_chan
->all_slots
);
1576 iop_chan
->common
.device
= dma_dev
;
1577 list_add_tail(&iop_chan
->common
.device_node
, &dma_dev
->channels
);
1579 if (dma_has_cap(DMA_MEMCPY
, dma_dev
->cap_mask
)) {
1580 ret
= iop_adma_memcpy_self_test(adev
);
1581 dev_dbg(&pdev
->dev
, "memcpy self test returned %d\n", ret
);
1583 goto err_free_iop_chan
;
1586 if (dma_has_cap(DMA_XOR
, dma_dev
->cap_mask
) ||
1587 dma_has_cap(DMA_MEMSET
, dma_dev
->cap_mask
)) {
1588 ret
= iop_adma_xor_val_self_test(adev
);
1589 dev_dbg(&pdev
->dev
, "xor self test returned %d\n", ret
);
1591 goto err_free_iop_chan
;
1594 if (dma_has_cap(DMA_PQ
, dma_dev
->cap_mask
) &&
1595 dma_has_cap(DMA_PQ_VAL
, dma_dev
->cap_mask
)) {
1596 #ifdef CONFIG_MD_RAID6_PQ
1597 ret
= iop_adma_pq_zero_sum_self_test(adev
);
1598 dev_dbg(&pdev
->dev
, "pq self test returned %d\n", ret
);
1600 /* can not test raid6, so do not publish capability */
1601 dma_cap_clear(DMA_PQ
, dma_dev
->cap_mask
);
1602 dma_cap_clear(DMA_PQ_VAL
, dma_dev
->cap_mask
);
1606 goto err_free_iop_chan
;
1609 dev_printk(KERN_INFO
, &pdev
->dev
, "Intel(R) IOP: "
1610 "( %s%s%s%s%s%s%s)\n",
1611 dma_has_cap(DMA_PQ
, dma_dev
->cap_mask
) ? "pq " : "",
1612 dma_has_cap(DMA_PQ_VAL
, dma_dev
->cap_mask
) ? "pq_val " : "",
1613 dma_has_cap(DMA_XOR
, dma_dev
->cap_mask
) ? "xor " : "",
1614 dma_has_cap(DMA_XOR_VAL
, dma_dev
->cap_mask
) ? "xor_val " : "",
1615 dma_has_cap(DMA_MEMSET
, dma_dev
->cap_mask
) ? "fill " : "",
1616 dma_has_cap(DMA_MEMCPY
, dma_dev
->cap_mask
) ? "cpy " : "",
1617 dma_has_cap(DMA_INTERRUPT
, dma_dev
->cap_mask
) ? "intr " : "");
1619 dma_async_device_register(dma_dev
);
1625 dma_free_coherent(&adev
->pdev
->dev
, plat_data
->pool_size
,
1626 adev
->dma_desc_pool_virt
, adev
->dma_desc_pool
);
1633 static void iop_chan_start_null_memcpy(struct iop_adma_chan
*iop_chan
)
1635 struct iop_adma_desc_slot
*sw_desc
, *grp_start
;
1636 dma_cookie_t cookie
;
1637 int slot_cnt
, slots_per_op
;
1639 dev_dbg(iop_chan
->device
->common
.dev
, "%s\n", __func__
);
1641 spin_lock_bh(&iop_chan
->lock
);
1642 slot_cnt
= iop_chan_memcpy_slot_count(0, &slots_per_op
);
1643 sw_desc
= iop_adma_alloc_slots(iop_chan
, slot_cnt
, slots_per_op
);
1645 grp_start
= sw_desc
->group_head
;
1647 list_splice_init(&sw_desc
->tx_list
, &iop_chan
->chain
);
1648 async_tx_ack(&sw_desc
->async_tx
);
1649 iop_desc_init_memcpy(grp_start
, 0);
1650 iop_desc_set_byte_count(grp_start
, iop_chan
, 0);
1651 iop_desc_set_dest_addr(grp_start
, iop_chan
, 0);
1652 iop_desc_set_memcpy_src_addr(grp_start
, 0);
1654 cookie
= iop_chan
->common
.cookie
;
1659 /* initialize the completed cookie to be less than
1660 * the most recently used cookie
1662 iop_chan
->completed_cookie
= cookie
- 1;
1663 iop_chan
->common
.cookie
= sw_desc
->async_tx
.cookie
= cookie
;
1665 /* channel should not be busy */
1666 BUG_ON(iop_chan_is_busy(iop_chan
));
1668 /* clear any prior error-status bits */
1669 iop_adma_device_clear_err_status(iop_chan
);
1671 /* disable operation */
1672 iop_chan_disable(iop_chan
);
1674 /* set the descriptor address */
1675 iop_chan_set_next_descriptor(iop_chan
, sw_desc
->async_tx
.phys
);
1677 /* 1/ don't add pre-chained descriptors
1678 * 2/ dummy read to flush next_desc write
1680 BUG_ON(iop_desc_get_next_desc(sw_desc
));
1682 /* run the descriptor */
1683 iop_chan_enable(iop_chan
);
1685 dev_printk(KERN_ERR
, iop_chan
->device
->common
.dev
,
1686 "failed to allocate null descriptor\n");
1687 spin_unlock_bh(&iop_chan
->lock
);
1690 static void iop_chan_start_null_xor(struct iop_adma_chan
*iop_chan
)
1692 struct iop_adma_desc_slot
*sw_desc
, *grp_start
;
1693 dma_cookie_t cookie
;
1694 int slot_cnt
, slots_per_op
;
1696 dev_dbg(iop_chan
->device
->common
.dev
, "%s\n", __func__
);
1698 spin_lock_bh(&iop_chan
->lock
);
1699 slot_cnt
= iop_chan_xor_slot_count(0, 2, &slots_per_op
);
1700 sw_desc
= iop_adma_alloc_slots(iop_chan
, slot_cnt
, slots_per_op
);
1702 grp_start
= sw_desc
->group_head
;
1703 list_splice_init(&sw_desc
->tx_list
, &iop_chan
->chain
);
1704 async_tx_ack(&sw_desc
->async_tx
);
1705 iop_desc_init_null_xor(grp_start
, 2, 0);
1706 iop_desc_set_byte_count(grp_start
, iop_chan
, 0);
1707 iop_desc_set_dest_addr(grp_start
, iop_chan
, 0);
1708 iop_desc_set_xor_src_addr(grp_start
, 0, 0);
1709 iop_desc_set_xor_src_addr(grp_start
, 1, 0);
1711 cookie
= iop_chan
->common
.cookie
;
1716 /* initialize the completed cookie to be less than
1717 * the most recently used cookie
1719 iop_chan
->completed_cookie
= cookie
- 1;
1720 iop_chan
->common
.cookie
= sw_desc
->async_tx
.cookie
= cookie
;
1722 /* channel should not be busy */
1723 BUG_ON(iop_chan_is_busy(iop_chan
));
1725 /* clear any prior error-status bits */
1726 iop_adma_device_clear_err_status(iop_chan
);
1728 /* disable operation */
1729 iop_chan_disable(iop_chan
);
1731 /* set the descriptor address */
1732 iop_chan_set_next_descriptor(iop_chan
, sw_desc
->async_tx
.phys
);
1734 /* 1/ don't add pre-chained descriptors
1735 * 2/ dummy read to flush next_desc write
1737 BUG_ON(iop_desc_get_next_desc(sw_desc
));
1739 /* run the descriptor */
1740 iop_chan_enable(iop_chan
);
1742 dev_printk(KERN_ERR
, iop_chan
->device
->common
.dev
,
1743 "failed to allocate null descriptor\n");
1744 spin_unlock_bh(&iop_chan
->lock
);
1747 MODULE_ALIAS("platform:iop-adma");
1749 static struct platform_driver iop_adma_driver
= {
1750 .probe
= iop_adma_probe
,
1751 .remove
= __devexit_p(iop_adma_remove
),
1753 .owner
= THIS_MODULE
,
1758 static int __init
iop_adma_init (void)
1760 return platform_driver_register(&iop_adma_driver
);
1763 static void __exit
iop_adma_exit (void)
1765 platform_driver_unregister(&iop_adma_driver
);
1768 module_exit(iop_adma_exit
);
1769 module_init(iop_adma_init
);
1771 MODULE_AUTHOR("Intel Corporation");
1772 MODULE_DESCRIPTION("IOP ADMA Engine Driver");
1773 MODULE_LICENSE("GPL");